Jump to content

Recommended Posts

I read through the USWNR rankings for statistics grad schools and don't know if I understand the ordering. Ignoring biostatistics, the rankings look like:

  1. Stanford
  2. Berkeley
  3. Harvard
  4. University of Chicago
  5. Carnegie Mellon
  6. University of Washington
  7. Duke
  8. University of Michigan
  9. Wharton
  10. Columbia
  11. NC State

As I am applying to Ph.D. programs for the fall 2021 cycle, I researched the departments. I will end up applying to almost all of these programs but wanted to ask the forum if my viewpoints make sense.

From what I have gathered, I do not get how Harvard is ranked so highly. The department is very small and largely focuses on two things, Bayesian inference and sampling, or design of experiments and causal inference. There seems to be one faculty member working on machine learning and given Ryan Adam's departure, it looks like they do not have much breadth at all in research areas. They do seem to have a large number of faculty in the biostatistics department, but that is a separate program all together (though I think it would be possible to work with these faculty quite easily).

Similar thoughts regarding University of Chicago. The department looks to be theory focused and more like an applied math department rather than a statistics department. They have a large fraction of faculty working on probability, PDEs, mathematical finance, etc. Only a small subset work on statistical methodology and theory. Given John Lafferty's departure, it also looks like they do not have much depth in ML.

If one were to re-do these rankings with a focus on ML, would it be correct to say, the top four should look more like Stanford, Berkeley, Carnegie Mellon, University of Washington?

 

 

Link to post
Share on other sites

Well, the rankings are not based on machine learning expertise, so I don't think it's fair to criticize the department rankings because they don't have a certain research focus.

That being said, I think a lot of people would agree about Chicago. They lost Lafferty recently and the focus of their department is pretty bizarre.  But they still have some really solid established people, some great stat gen people like Stephens.  I would say you should take all the US News rankings with a plus or minus 10, at least, and rankings should not be the only consideration, especially within the top 20.

I would disagree with your assessment of Harvard though. Meng, Murphy, Kou, Imai, Xihong Lin, and more. Most of these are not working just in those two areas you mentioned and are extremely famous. They are as good as anyone you will find anywhere.  Murphy is MacArthur genius grant winner who focuses on reinforcement learning.

I don't think it's a very useful exercise to rank these places for ML. Stanford has all the LASSO/HD people, Berkeley has Jordan, Washington has Witten, CMU has Wasserman, Harvard has Murphy, Michigan and NCSU have plenty of people.  Also, ML is such a broad label that I don't think it's an extremely useful thing to be thinking about when choosing a stats department - LASSO and reinforcement learning and Bayesian non-parametrics are are "ML" topics but don't really have anything more in common than any other two areas of stats. Find professors whose work is interesting to you and apply there. 

Link to post
Share on other sites

I agree with @bayessays and @statistican. IIRC, the rankings are determined by surveys sent out to professors in stats departments. USNWR averages the scores and ranks departments from there. UChicago has a great reputation among faculty, therefore it's high in the rankings. A lot of things can change between surveys (which I believe happen every 4-5 years), and a lot of important info is imperceptible to people outside a department (e.g. your prospective advisor plans to retire soon). So within 10 or so places, take the rankings with a grain of salt, like bayessays said. 

W.r.t. to ML research, it seems like most departments have been trying hard to increase their presence in ML. However, ML is only one part of modern stats research. ML itself is broad, so if you're dead-set on ML, I would encourage you to consider how connected the stats department is to the CS department. That could augment you choices of research and advisors.

Link to post
Share on other sites
2 hours ago, Bayequentist said:

On a side note, does anyone know when will US News refresh the rankings for stat/biostat?

I believe they rank them every 4 years. The last ones just came out in 2018, so I'd expect new ones in 2022.  The biggest changes are the additions of new programs that were previously not ranked.

Link to post
Share on other sites

What do you guys think about JHU’s applied math and stat department, its ranked a little lower , I think it’s at 31, is it a good dept for HD and financial statistics? Also is it really that much easier to get into it given the low rank?

Link to post
Share on other sites
35 minutes ago, movingtostats said:

What do you guys think about JHU’s applied math and stat department, its ranked a little lower , I think it’s at 31, is it a good dept for HD and financial statistics? Also is it really that much easier to get into it given the low rank?

Not sure about HD/financial stats, but they have some solid people.  One of their alums just got a tenure-track job at Pitt, but I think one of his two advisors just left to be a prof at NCSU - your best bet is probably his other advisor, Carey Priebe.  Geman invented the Gibbs sampler, but I'm not sure that he still advises students.  The financial math people are not traditional statisticians but may be fine depending on your goals and interests.  Then they have a few people who work in statistics-adjacent things like computational imaging.  Honestly, because they have so few statisticians, it probably doesn't compare to other programs in the 30s-low 40s, in my opinion, but they have had some recent alums with success.

Link to post
Share on other sites

If you are interested in machine learning/high-dimensional statistics, don't overlook some of the top biostatistics programs. For example, University of Washington Biostatistics has some excellent faculty for this, e.g. Witten and Shojaie. PhD graduates from good Biostatistics PhD programs that emphasize both theory and clinical applications are able to get jobs in traditional statistics departments (for example, Edward Kennedy at CMU Statistics has a PhD in Biostatistics and publishes a lot in the top stat journals).

Link to post
Share on other sites
17 hours ago, bayessays said:

Not sure about HD/financial stats, but they have some solid people.  One of their alums just got a tenure-track job at Pitt, but I think one of his two advisors just left to be a prof at NCSU - your best bet is probably his other advisor, Carey Priebe.  Geman invented the Gibbs sampler, but I'm not sure that he still advises students.  The financial math people are not traditional statisticians but may be fine depending on your goals and interests.  Then they have a few people who work in statistics-adjacent things like computational imaging.  Honestly, because they have so few statisticians, it probably doesn't compare to other programs in the 30s-low 40s, in my opinion, but they have had some recent alums with success.

I see, yes they have quite a few guys working in probability/ stochastic processes as well I think. Their department seems more like an operations research and applied math dept rolled into one. How does the difficulty level of getting into their programme compare with that of other programs, I’m guessing it’s maybe rougly the same as that of duke’s? 

Link to post
Share on other sites
9 hours ago, Stat Postdoc Soon Faculty said:

If you are interested in machine learning/high-dimensional statistics, don't overlook some of the top biostatistics programs. For example, University of Washington Biostatistics has some excellent faculty for this, e.g. Witten and Shojaie. PhD graduates from good Biostatistics PhD programs that emphasize both theory and clinical applications are able to get jobs in traditional statistics departments (for example, Edward Kennedy at CMU Statistics has a PhD in Biostatistics and publishes a lot in the top stat journals).

One thing I don’t understand is that why is it so much easier to get into top biostat programs than top stats programs when most people say that the two depts do mostly the same things.

Link to post
Share on other sites
4 minutes ago, movingtostats said:

One thing I don’t understand is that why is it so much easier to get into top biostat programs than top stats programs when most people say that the two depts do mostly the same things.

It is actually extremely difficult for international students to get into top biostat programs. The funding structure there favors domestic students (often times, funding in biostat comes from things like NIH training grants, whereas Statistics departments will fund a lot of their students through TAship in undergrad statistics courses; there aren't usually a lot of undergrad biostatistics courses). There are, in general, fewer American students looking to get PhDs in Statistics (there are a lot more pursuing only Masters degrees). So that is partly why it is somewhat less competitive for American applicants to PhD programs.

Link to post
Share on other sites
2 minutes ago, movingtostats said:

One thing I don’t understand is that why is it so much easier to get into top biostat programs than top stats programs when most people say that the two depts do mostly the same things.

The top 3/4 biostat programs (Harvard, Hopkins, Washington, UNC) are the only ones you can really compare to top stat programs. I would say that the few top theoretical programs (Stanford, Penn, maybe some others) are harder to get into than Harvard/Hopkins, but I'm not sure there's a huge difficulty gap in admissions any more between those programs and a place like Michigan/CMU.  Most researchers are doing roughly the same research in all these top departments, so the difference comes into play at the extremes.  A place like Stanford is going to have some extremely theoretical people that a biostat department would never have, so they'll want some people who can handle that and who can take the theoretical probability courses.  A biostat department will have some people doing extremely applied work that probably wouldn't be done in a stat department, and the coursework is generally easier (at non-top programs) so they can be a little more forgiving.  

Link to post
Share on other sites
3 hours ago, bayessays said:

The top 3/4 biostat programs (Harvard, Hopkins, Washington, UNC) are the only ones you can really compare to top stat programs. I would say that the few top theoretical programs (Stanford, Penn, maybe some others) are harder to get into than Harvard/Hopkins, but I'm not sure there's a huge difficulty gap in admissions any more between those programs and a place like Michigan/CMU.  Most researchers are doing roughly the same research in all these top departments, so the difference comes into play at the extremes.  A place like Stanford is going to have some extremely theoretical people that a biostat department would never have, so they'll want some people who can handle that and who can take the theoretical probability courses.  A biostat department will have some people doing extremely applied work that probably wouldn't be done in a stat department, and the coursework is generally easier (at non-top programs) so they can be a little more forgiving.  


I see, so I guess biostats is maybe not the best option in general for somebody who wants to work only in theory, but it is probably just as good as the stats depts if somebody is interested in applied or a mixture of applied and theory.

Link to post
Share on other sites
3 hours ago, Stat Postdoc Soon Faculty said:

It is actually extremely difficult for international students to get into top biostat programs. The funding structure there favors domestic students (often times, funding in biostat comes from things like NIH training grants, whereas Statistics departments will fund a lot of their students through TAship in undergrad statistics courses; there aren't usually a lot of undergrad biostatistics courses). There are, in general, fewer American students looking to get PhDs in Statistics (there are a lot more pursuing only Masters degrees). So that is partly why it is somewhat less competitive for American applicants to PhD programs.


Ohh I had now idea that the funding worked in such a different way for biostat programs.

Link to post
Share on other sites
30 minutes ago, movingtostats said:


I see, so I guess biostats is maybe not the best option in general for somebody who wants to work only in theory, but it is probably just as good as the stats depts if somebody is interested in applied or a mixture of applied and theory.

To a large extent, this is true, but I want to be very clear about some terminology here which most people only become very familiar with after being in a graduate program, so if you're already aware of this distinction, I don't mean to be talking down.  There are broadly three types of research, applied, methodological, and theoretical.

Applied research is exactly what it sounds like - you are analyzing some data set and want to answer a scientific question.  There is more opportunity to do this type of research in a biostatistics program because there are so many biomedical collaborations.  You might be funded by working on applied research.  But nobody receives a PhD in statistics because they just did applied research.

The majority of what would be considered statistics research is methodological, creating or slightly extending methods to apply to new types of data or in unique situations.  There will probably be some application in the paper, but the focus is on the method, how it performs, etc.  Some of these papers are very "theoretical" in the laymen's sense that they use a lot of math and have some proofs. Scroll through the following papers, quickly look at the math, and decide if this is theoretical enough for you: 

https://arxiv.org/pdf/1910.04291.pdf

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5560424/

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3285536/

This type of research is done in all the top biostatistics departments.  Even biostatistics programs ranked 15th have people doing stuff like this.  And most people in statistics programs are doing research exactly like this too - this is where the large area of overlap between statistics and biostatistics comes from.

When people are talking about theoretical statistics, they are generally talking about *extremely* theoretical, math-heavy stuff you would find in a journal like Annals of Statistics.  Compare the above papers to the two below:
https://arxiv.org/pdf/1705.07382.pdf

https://arxiv.org/pdf/2003.05599.pdf

You will, generally, not find much research like the above in a biostatistics department, outside of 1 or 2 professors at top departments.  Most professors in stats departments are not even doing research like this.

In summary, think of how theoretical of research you really want to do.  If the first 3 papers roughly fit that, a biostatistics department will be fine.  In my experience, the overwhelming majority of students, and especially domestic students, do not want to work on the type of extremely theoretical problems anyways.

Link to post
Share on other sites
49 minutes ago, bayessays said:

To a large extent, this is true, but I want to be very clear about some terminology here which most people only become very familiar with after being in a graduate program, so if you're already aware of this distinction, I don't mean to be talking down.  There are broadly three types of research, applied, methodological, and theoretical.

Applied research is exactly what it sounds like - you are analyzing some data set and want to answer a scientific question.  There is more opportunity to do this type of research in a biostatistics program because there are so many biomedical collaborations.  You might be funded by working on applied research.  But nobody receives a PhD in statistics because they just did applied research.

The majority of what would be considered statistics research is methodological, creating or slightly extending methods to apply to new types of data or in unique situations.  There will probably be some application in the paper, but the focus is on the method, how it performs, etc.  Some of these papers are very "theoretical" in the laymen's sense that they use a lot of math and have some proofs. Scroll through the following papers, quickly look at the math, and decide if this is theoretical enough for you: 

https://arxiv.org/pdf/1910.04291.pdf

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5560424/

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3285536/

This type of research is done in all the top biostatistics departments.  Even biostatistics programs ranked 15th have people doing stuff like this.  And most people in statistics programs are doing research exactly like this too - this is where the large area of overlap between statistics and biostatistics comes from.

When people are talking about theoretical statistics, they are generally talking about *extremely* theoretical, math-heavy stuff you would find in a journal like Annals of Statistics.  Compare the above papers to the two below:
https://arxiv.org/pdf/1705.07382.pdf

https://arxiv.org/pdf/2003.05599.pdf

You will, generally, not find much research like the above in a biostatistics department, outside of 1 or 2 professors at top departments.  Most professors in stats departments are not even doing research like this.

In summary, think of how theoretical of research you really want to do.  If the first 3 papers roughly fit that, a biostatistics department will be fine.  In my experience, the overwhelming majority of students, and especially domestic students, do not want to work on the type of extremely theoretical problems anyways.

This is amazing!!! Don’t worry, you’re not talking down. If I’m being honest I didn’t find the first three papers to be theoretical enough, and the last two might have been a tad too theoretical , I think I am looking for something in the left neighborhood of the last two papers which I guess can be found more in the stats depts but then again stats depts are too difficult to get into so maybe I’ll have to go the biostats route. But I highly doubt my capabilities being from a non math background, I’m still trying to shift into statistics and will try to get an MS and see how it goes from there. Btw if you’ve got time I’d love for you to tell me which schools I should apply to for an MS, here is the post I had made a while ago https://forum.thegradcafe.com/topic/123426-profile-evaluation-ms-applicant/

 

Link to post
Share on other sites
2 minutes ago, movingtostats said:

This is amazing!!! Don’t worry, you’re not talking down. If I’m being honest I didn’t find the first three papers to be theoretical enough, and the last two might have been a tad too theoretical , I think I am looking for something in the left neighborhood of the last two papers which I guess can be found more in the stats depts but then again stats depts are too difficult to get into so maybe I’ll have to go the biostats route. But I highly doubt my capabilities being from a non math background, I’m still trying to shift into statistics and will try to get an MS and see how it goes from there. Btw if you’ve got time I’d love for you to tell me which schools I should apply to for an MS, here is the post I had made a while ago https://forum.thegradcafe.com/topic/123426-profile-evaluation-ms-applicant/

 

Those last 2 papers are on the very extreme end of theoretical, but if you want to mainly do stuff more theoretical than the first 3, you're probably right that a stats department is better.  You can do that type of research at even a low-ranked statistics department, but you won't find stuff more theoretical than the first 3 outside the top biostat departments.

I don't have much to add to the MS recommendations.  Apply to a wide range of schools.  Most schools with a PhD program would be good.  Wake Forest has an impressive program where you can get good training and write a thesis.

Link to post
Share on other sites
4 minutes ago, bayessays said:

Those last 2 papers are on the very extreme end of theoretical, but if you want to mainly do stuff more theoretical than the first 3, you're probably right that a stats department is better.  You can do that type of research at even a low-ranked statistics department, but you won't find stuff more theoretical than the first 3 outside the top biostat departments.

I don't have much to add to the MS recommendations.  Apply to a wide range of schools.  Most schools with a PhD program would be good.  Wake Forest has an impressive program where you can get good training and write a thesis.

Thanks a lot, I always thought that there had to be a reason for the (relatively)low math requirements at biostat programs, now it makes sense. Thanks for the advice about wake forest, I’ll definitely check it out. I was also thinking about chicago and duke but they’re very tough to get into. One last thing do you think I should give the mGRE to show my mathematical ability(assuming I can do well on it) or is it a waste of time?

Link to post
Share on other sites

I personally don't think it's worthwhile for a MS degree.  I suppose there is a chance someone would look at it, but even Chicago doesn't recommend MS applicants to take the subject test.  It would be a shame to spend all that time studying for a test that won't end up helping you at all.

Link to post
Share on other sites
On 4/22/2020 at 9:09 PM, bayessays said:

The top 3/4 biostat programs (Harvard, Hopkins, Washington, UNC) are the only ones you can really compare to top stat programs. I would say that the few top theoretical programs (Stanford, Penn, maybe some others) are harder to get into than Harvard/Hopkins, but I'm not sure there's a huge difficulty gap in admissions any more between those programs and a place like Michigan/CMU.  Most researchers are doing roughly the same research in all these top departments, so the difference comes into play at the extremes.  A place like Stanford is going to have some extremely theoretical people that a biostat department would never have, so they'll want some people who can handle that and who can take the theoretical probability courses.  A biostat department will have some people doing extremely applied work that probably wouldn't be done in a stat department, and the coursework is generally easier (at non-top programs) so they can be a little more forgiving.  

What makes Michigan/CMU different from Stanford/Penn? Are you saying that the admissions difficulty is not that different from top biostats programs for Michigan/CMU because they are less theoretical? I know that Stanford is considered a league of its own, but my research advisor (assistant prof at a top 10 stats program) told me he who would rank Michigan/CMU above Penn hence I am surprised you would say they Penn is harder to get into than Harvard/Hopkins.

Link to post
Share on other sites

Penn is a very small program; I think their cohort size is 5 a year so their department might not be as strong as say, Michigan or CMU, but getting in might be as or more difficult because CMU and Michigan have cohort sizes of about double that. Also in general, Ivies are harder to get into relative to their departmental strength due to their name outside of statistics.

Link to post
Share on other sites

Above is correct. Penn has 31 PhD students while Michigan and CMU 55-70 each.  Penn is a very strong department, I would not put it below Michigan and CMU, but they are comparably strong and it would be reasonable to go to any of them if you had the choice.  Penn is often seen as a very theoretical program and places well in academia. 

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...

Important Information

By using this site, you agree to our Terms of Use and Privacy Policy.